CS528

Power/Energy Aware Cloud System

Design

A. Sahu

Dept of CSE, IIT Guwahati

Outline

• Power/Energy Consumption Model

• Power Aware Computing

• Thermal Aware Computing

• Power Aware Scheduling in Cloud

• Migration and Management

Data Center Power Consumption

• Currently it is estimated that servers consume

0.5% of the world’s total electricity usage.

– Closer to 1.2% when data center systems are factored

into the equation.

• Server energy demand doubles every 4-6 years.

• This results in large amounts of CO2 produced by

burning fossil fuels.

• What if we could reduce the energy used with

minimal performance impact?

3

Percentage of Power Consumption in

DC

Motivation for Green Data Centers

• Economic

– New data centers run on the Megawatt

scale, requiring millions of dollars to

operate.

– Recently institutions are looking for new

ways to reduce costs, no more “blank

checks.”

– Many facilities are at their peak

operating envelope, and cannot expand

without a new power source.5

Motivation for Green Data Centers

• Environmental

– 70% of the U.S. energy sources are fossil

fuels. In India 66% is from Coal Plant

(NTPC)

– 2.8 billion tons of CO2 emitted each year

from U.S. power plants.

– Sustainable energy sources are not ready.

– Need to reduce energy dependence until

a more sustainable energy source is

deployed. 6

Green Cloud Goal Shift:

“performance” ���� “energy efficiency”

• As energy costs are increasing while availability

dwindles

• Need to shift focus optimising data center Resource

management

– From pure performance alone to optimising for

energy efficiency

– While maintaining high service level performance.

• Green Cloud computing model that achieves

– not only efficient processing and utilisation of computing

infrastructure,

– but also minimise energy consumption.

Green Computing• In the past 15-20 years of supercomputers

o performance has doubled > 3000 timeso performance per watt has doubled 300 timeso performance per square foot has doubled 65 times

Revenue

Power

Consumption

Green Cloud Computing

Internet Service Provider

Routers

Internet

CloudCloudDatacenter A

End User

CloudCloudDatacenter B

Cloud Cloud Datacenter C

Cloud Usage Model

Cloud Usage Model

Cloud

Datacenter

Datacenter

LAN and Gateway

router

(Network Devices)

VM and Storage

(Server)

Air Conditioning, and Chiller (Cooling Devices)

UPS, PDU, lighting

(Electrical Devices)

Cloud

users

Cloud Computing

• Features of Clouds

– Scalable, Enhanced Quality of Service (QoS)

– Specialized and Customized, Cost Effective

– Simplified User Interface

12Compute Node

Compute

Node

Compute

Node

File

Server

Head

Node

VM VM VM VM VM VM

Execute job in a VM

Start a VM

Job

System model

• Considered IaaS paradigm

• Consists of m homogenous hosts (m is large)

vt1

vt2

vtk

PM

PM

PM

PMm

2

1

3

t2

t1

t n

Cloud System

k VM types m PMs

n Tasks

• Every Task comes with

•Execution time (ei), deadline (di)

•CPU requirement (ci), memory requirement (mi)

•and any other

• SLA (Service Level Agreement ) Violation

•SLAV

•If the task do not get require amount CPU, memory

•If competition time extend the deadline

•Cloud service provide may need to pay penalty

System Model

Green Cloud Computing Framework

Green Cloud

Framework

VM

Controls

Scheduling

Power

Aware

Thermal

Aware

Management

VM

Design

VM

Image

Design

Migr-

ation

Dynamic

Shutdown

Data Center

Design

AC &

lation

AC &

Recircu-

lation

Server &

Placement

Server &

Rack

Placement

Green Cloud Framework (cont.)

• Goal : Maximize performance per watt in

Cloud – VM Scheduling

– VM Image Management

– Data Center Design

• Scheduling – Placement within cloud infrastructure

– Energy use of server equipment

– datacenter temperature important

Green Cloud Framework (cont.)

• Image Management – Small Size

– Few unnecessary processes/services

– Migration

– Dynamic Shutdown

• Data Center Design – More efficient A/C, power supplies

– Hot and cold aisles

– Utilizing external cooling

Virtual Machine Scheduling

• Power-Aware Scheduling (PAS)– Minimize total power used by servers

– Power to servers is the larger cost

• Thermal-Aware Scheduling (PAS)– Minimize overall temperature

– Reduces energy used for cooling

Virtual Machine Scheduling

Power consumption curve on an Intel Core i7 920 Server

(4 cores, 8 virtual cores with Hyperthreading)

90

110

130

150

170

190

0 1 2 3 4 5 6 7 8

Wa

tts

Number of Processing Cores

Power Aware (PA) Computing

• Objective of PA computing/communications is

– To improve power management and

consumption

– Using the awareness of power consumption of

devices.

• Power consumption is most important considerations

– In mobile devices due to limitation battery life.

Power Aware Computing

• System level power management

• Recent devices support multiple power modes.

– CPU, disk, communication links, etc.

• Resource Management and Scheduling Systems

– Can use these multiple power modes

– To reduce the power consumption.

Real Life Issue: Inverter AC

• Inverter AC vs Non-Inverter AC

• Non-Inverter AC : Run fast and rest

• Non-Inverter AC: switch-of and switch-on

mode

– Sound, Fan on-off

• Inverter AC : Quit and required

– Run at required speed : Fun to compare with EMI

– Quieter than a mosquito

Real Life Issue: Inverter AC• Eco Friendly, less power consumption

• Makes little sound, Efficient Cooling/Heating

• No Voltage Fluctuation caused by compressor

• Can be run on solar panels

DPM vs DVFS

• Inverter AC vs Non-Inverter AC

• Non-Inverter AC : Run fast and rest

• DPM : switch-of and switch-on mode

– Sound, Fan on-off

• DVFS : Quit and required mode

– Quieter than a mosquito

– Run at required speed

DVFS

• Dynamic Voltage and Frequency Scaling – Intel SpeedStep

– AMD PowerNow

• Started in laptops and mobile devices

• Now used in servers

DVS (Dynamic Voltage Scaling)• Reducing the dynamic energy consumption

– By lowering the supply voltage at the cost of

performance degradation

• Recent processors support such ability

– To adjust the supply voltage dynamically.

• The dynamic energy consumption

– α * Vdd2 * Ncycle

Vdd : the supply voltage, Ncycle : the number of

clock cycle

DVS (Dynamic Voltage Scaling)

10 msec 25 msec

deadlinePower

5.02

(a) Supply voltage = 5.0 V

10 msec 25 msec

deadlinePower

2.02

(b) Supply voltage = 2.0 V

DVFS-based Power Aware Scheduling :

Motivation

• Develop Resource Management and Scheduling Algorithms

– That aim at minimizing the energy

consumption

– At the same meet the job deadline.

• Exploit industrial move towards

– Utility Model/SLA-based Resource Allocation

for Cloud Computing